Re: [PATCH v2 0/6] ARM: shmobile: sh73a0: DT PM domain support

[Simon Horman <horms@verge.net.au>]

On Thu, Jan 15, 2015 at 10:32:23AM +0100, Geert Uytterhoeven wrote:
> Hi Simon,
> 
> On Thu, Jan 15, 2015 at 12:53 AM, Simon Horman <horms@verge.net.au> wrote:
> > On Wed, Jan 14, 2015 at 01:11:18PM +0100, Geert Uytterhoeven wrote:
> >> This patch series enables DT support for PM domains on the Renesas
> >> SH-Mobile AG5 (sh73a0) SoC.
> >>
> >> This series builds further on the DT PM Domain support for R-Mobile A1
> >> (r8a7740).  Due to the similarity of the SYSC System-Controller on the
> >> various SH-Mobile/R-Mobile SoCs, and the abstraction of PM domains in
> >> DT, the same driver can handle sh73a0 after some small modifications,
> >> without changing the DT bindings.
> >>
> >> Compared to r8a7740, the only significant change is the move of the
> >> memory-controller(s) to separate PM domains on sh73a0 (and also on
> >> R-Mobile APE6 (r8a73a4)). Hence a provision must be made to not turn off
> >> these PM domains inadvertently, cfr. the series "[PATCH 0/4] ARM:
> >> shmobile: Add DT support for memory controllers" I've just sent.
> >>
> >> This series has been sent before, as part of the series "[PATCH RFC
> >> 0/7] ARM: shmobile: sh73a0: DT PM domain support"
> >> (https://lkml.org/lkml/2014/11/19/404).
> >> Changes since v1 (more detailed changelogs in the individual patches):
> >>   - Factored out addition of the memory-controller in a separate
> >>     series.
> >>
> >> Dependencies:
> >>   1. Patches 1-4 (DT binding doc, C code) depend on:
> >>        a. The R-Mobile DT PM Domain work for r8a7740, queued in tag
> >>           renesas-dt-pm-for-v3.20 of Simon's repository,
> >
> > In response to a request from Olof renesas-dt-pm-for-v3.20 has been
> > withdrawn and the patches split between the soc-for-v3.20 and dt-for-v3.20
> > branches. I am assuming that the patches that 1-4 depend on are
> > now all in the soc-for-v3.20 branch. Could you check?
> 
> OK, soc-for-v3.20 now.
> 
> >>        b. "ARM: shmobile: R-Mobile: Fix DT refcount bugs in PM domain
> >>          code"
> >>          (http://www.spinics.net/lists/arm-kernel/msg391084.html),
> >
> > FWIW, I have squashed that patch into the patch that adds the problem
> > and the result is in the soc-for-v3.20. I will to push that result
> > a little later today.
> 
> Thanks!
> 
> >>   2. Patch 5 (dtsi) depends on:
> >>        a. The multiplatform work for sh73a0, queued in branch
> >>           sh73a0-multiplatform-for-v3.20 of Simon's repository,
> >>        b. Series "[PATCH 0/4] ARM: shmobile: Add DT support for memory
> >>           controllers"
> >>         (http://marc.info/?l=linux-sh&m=142123399414888).
> >
> > This may be awkward to get into v3.20 unless we can adopt a looser approach
> > to the dependencies.
> >
> > To that end I suggest that it could be queued up in the branch that has the
> > multiplatform work (a) but not the memory controller work (b) as the latter
> > seems to more naturally fit into the dt branch. What I am hoping that such
> > an approach would lead to code that compiles and boots in both branches.
> > And that new functionality becomes available when the branched are combined
> > (e.g. in v3.20-rcX).
> 
> "ARM: shmobile: sh73a0 dtsi: Add PM domain support" relies on the clocks
> properties already being present on all device nodes (incl. the Bus State
> Controller), hence it depends on sh73a0-multiplatform-for-v3.20.
> "ARM: shmobile: sh73a0 dtsi: Add PM domain support" adds a power-domains
> property to the memory-controller node, so it depends on that, too.
> 
> Leaving the memory-controller part out makes it unbootable, as the
> memory-controller will be powered down when PM domains become
> activated. Yes, hardware power domains are even more fun than (common)
> clocks when trying to split the commits across multiple branches ;-)
> 
> > If this is not possible then I can merge the pre-requisites and put
> > patch 5 on top. But I feel that the ARM SoC maintainers would prefer
> > that I didn't do that unless it is absolutely necessary.
> 
> Merging dt-for-v3.20 into sh73a0-multiplatform-for-v3.20 before applying
> patch 5 seems like the only sensible solution.

Thanks, I understand. I'll see about making it so.

> As soon as that gets combined with soc-for-v3.20, the PM domains are
> activated, and patch 6 can be applied.

Got it.

> > In case you are wondering: branch gymnastics are not my favourite pass time :^)
> >
> >>   3. Patch 6 (drivers/sh) depends on all of the above.
> >
> > That part is clear enough :)
> >
> > My intention here is to send a pull request to Linus for this change
> > once all the other patches have hit his tree. With a bit of luck that
> > would be around v3.20-rc2.
> 
> OK, that would be appied to sh-drivers-for-v3.20, right?

Yes, thats right.

> Thanks a lot!
> My local patch queue has shrunk considerably during the last few days...

:^)